中高齡勞動狀態對心理健康影響之探討

Abstract

目的 本研究旨在探討台灣中高齡者勞動狀態改變對心理健康的因果效應。面對台灣於2025年邁入超高齡社會的勞動力緊縮壓力，政府積極推動延長中高齡就業之政策。然而，既有文獻對於勞動與健康的關係存在分歧，且兩者間具有強烈的雙向因果與內生性。本研究嘗試釐清退休是否實質損及中高齡者的心理健康，並檢視現行促進就業政策是否合乎中高齡者的健康福祉。 方法 本研究運用「台灣地區中老年身心社會生活狀況長期追蹤調查（TLSA）」2003年至2019年共五波的長期追蹤次級資料。在排除失業與再就業樣本後，建構包含3,366人、14,473筆資料的主樣本，以及858人、3,931筆資料在期間內由就業轉入退休的子樣本。為處理兩變數之內生性、未觀察個體異質性與隨時間變動的干擾因子（time-varying confounding），為求估計穩健性，本研究採用兩種不同計量策略，第一，採用聯立方程式模型（Simultaneous Equations Model），以完全訊息最大概似法（Full Information Maximum Likelihood）進行聯合估計; 第二，使用因果推論潛在結果框架，利用流行病學之g-formula進行反事實情境模擬（包含退休、就業、延後一期與延後至70歲退休），捕捉勞動狀態對心理健康之群體平均效果（Population Average Effect）。 結果 本研究發現，描述性統計顯示退休群體的平均憂鬱分數顯著高於就業群體。然而進一步透過聯立方程式模型分析發現，前期退休對當期的心理健康可能僅具有微弱的正向效果（降低憂鬱分數）、兩方程式之內生性存在，且高齡勞工因不健康產生預算需求，被迫在健康尚可時續留職場以賺取工資。最後，g-formula分析結果顯示，對整體主樣本施加「持續勞動」會顯著惡化心理健康，而「延後一期退休」與「持續退休」能顯著降低憂鬱分數，然而在子樣本群體中，所有退休介入情境可能因樣本限縮、模型配適與預期心理而皆未達統計顯著。 共病與性別的次族群反事實模擬進一步顯示，多重共病對心理健康的負面衝擊遠大於勞動政策的變動，若施加「共病歸零」的介入能大幅減輕群體的憂鬱軌跡。此外，若將男性的社經與勞動分佈規則套用於女性，亦能幾乎消弭中高齡女性偏高的憂鬱分數。 結論 本研究證實，退休或離開職場本身並非損害中高齡者心理健康的原因。過往橫斷面觀察到的高憂鬱風險，本質上可能由多重共病、女性結構性弱勢與低社經地位等潛藏的健康不平等所主導。對於真正經歷勞動狀態轉換的中高齡者而言，延長勞動年限的政策衝擊表現得較不敏感，且高度受到個體生理健康影響。據此，政府之中高齡就業促進政策應將職場健康促進、職務再設計與退休後之社會福利政策置於勞動政策之政策形成考量，方能落實具備尊嚴與世代正義的健康老化。

Objectives: This study investigates the dynamic causal effects of labor market transitions on mental health among older adults in Taiwan. Facing severe labor shortage pressures as Taiwan transitions into a super-aged society by 2025, the government has actively promoted policies to extend working lives. However, existing literature presents divergent views on the health-employment nexus, which is inherently complicated by strong bidirectional causality and endogeneity. This study clarifies whether retirement substantively deteriorates older adults’ mental health and examines whether current employment promotion policies align with their health and well-being. Methods: This study analyzed secondary data from five waves (2003–2019) of the Taiwan Longitudinal Survey on Aging (TLSA). After excluding unemployed and re-employed individuals, we constructed a main sample of 3,366 participants (14,473 observations) and a subsample of 858 participants (3,931 observations) who transitioned from employment to retirement during the follow-up period. To address endogeneity, unobserved individual heterogeneity, and time-varying confounding, we adopted a dual-methodological strategy: (1) a simultaneous equations model estimated via Full-Information Maximum Likelihood (Full Information Maximum Likelihood), and (2) the epidemiological G-formula under the potential outcomes framework to simulate counterfactual intervention scenarios (including Always Retire, Never Retire, Delay Retire, and Retire at 70) and capture Population Average Effects (PAEs). Results: Descriptive statistics indicated that retirees had significantly higher average depressive symptom scores than employed individuals. However, simultaneous equations model analyses revealed that prior retirement exerted only a weak protective effect (lowering depressive scores) on current mental health, confirmed the presence of endogeneity, and suggested retaining workplace where health-induced budget constraints forced older workers to remain employed while physically capable. Furthermore, g-formula simulations on the main sample demonstrated that ‘Never Retire’ significantly worsened mental health, whereas ‘Delay Retire’ and ‘Always Retire’ significantly reduced depressive scores. Conversely, all intervention scenarios yielded statistically null effects in the transition subsample, potentially driven by sample size reduction, model fitting discrepancies, and anticipation smoothing. Subgroup counterfactual simulations revealed that multimorbidity exerted a far more profound negative impact on mental health than labor policy shifts; simulating ‘zero comorbidities’ substantially flattened depressive trajectories. Moreover, applying male socioeconomic and labor distribution rules to females almost entirely eliminated the excess depressive risks observed among older women. Conclusions: This study demonstrates that retirement for individual is not the main reason behind deteriorating mental health in later life. The elevated depressive risks observed in cross-sectional analysis are primarily driven by underlying health inequalities, such as multimorbidity, structural gender disadvantages, and low socioeconomic status. For older adults genuinely undergoing labor market exit, policy-induced shifts in retirement timing exhibit low sensitivity and are heavily dominated by physical health depreciation. Consequently, policymakers should prioritize workplace health promotion, job redesign, and post-retirement social safety nets as prerequisites for labor extension policies to foster truly dignified and equitable active ageing.